Visual Cue Integration for Depth Perception

Abstract

The human visual system obtains information about the depth of an object from a large number of distinct cues. Cues to depth result from object rotation, observer motion, binocular vision in which the two eyes receive different patterns of light, texture gradients in retinal images, and many other factors. A remarkable feature of human perception is that we are not overwhelmed by this wealth of information. Instead, we seem to effortlessly integrate information provided by each cue into aunified percept that is highly accurate. Moreover, we do this in a wide variety of visual environments. It is important to note that no single cue is necessary for depth perception or dominates our perception of depth (Cutting and Vishton, 1995). Moreover, no individual cue has been demonstrated to be individually capable of supporting depth perception with the robustness and accuracy shown by human observers. Clearly, our remarkable abilities to perceive visual depth are based on our abilities to integrate information provided by a variety of depth cues. In this chapter, we examine two important aspects of visual cue integration for depth perception. First, we address the question of whether or not observers integrate information based onmultiple cues in an efficientmanner. This can be evaluated by examining the degree to which their cue integration strategies can be characterized as statistically optimal in a Bayesian sense. Next, we look at the role that visual learning may play in cue integration. We address the question of whether or not observers cue integration strategies for visual depth are adaptable in an experience-dependent manner.